In a significant leap forward for on-device generative AI, AMD has unveiled a new wave of performance-optimized AI models, developed in collaboration with Stability AI. These models, now available through the newly launched Amuse 3.0 Beta by Tensorstack.AI, bring up to 4.3x faster inference for image and video generation on AMD Radeon graphics cards and Ryzen AI processors.
Amuse 3.0 is engineered to deliver top-tier creative tools natively on AMD hardware, integrating AI performance and accessibility in a single, seamless platform.
Table of Contents
AMD Optimized Models: A New Era for AI on Radeon and Ryzen AI
The highlight of this release is the AMD-optimized versions of popular models like Stable Diffusion 3.5, now labeled with the “_amdgpu” suffix on Hugging Face. These models are specially tuned for AMD Radeon GPUs and Ryzen AI processors, boasting speedups of up to 4.3x on the Radeon RX 9070 XT and up to 3.3x on Ryzen AI MAX+ 395 series processors.
The performance gains are striking:
- Stable Diffusion 1.5 sees up to 4.3x faster generation on Radeon RX 9070 XT.
- SDXL 1.0 performance boosts by up to 3.1x.
- SD 3.5 Large Turbo, previously inaccessible to many, now runs smoothly on AMD Ryzen AI systems.
What’s New in Amuse 3.0
Amuse 3.0 goes beyond just speed. The new features are tailor-made for creators:
- High-quality Photo AI Filters using SDXL Lightning models with ControlNets.
- Draft-quality Video Diffusion and Restyle, allowing video-to-video AI transformations entirely offline.
- Over 100 new image generation models and fine-tunes, including popular releases from Stability AI and Black Forest Labs.
Users with AMD Ryzen AI systems featuring XDNA NPUs can now generate 4MP print-quality images locally in minutes, taking privacy and performance to new heights.
Recommended Systems for the Best Experience
To maximize the benefits of these new models, AMD suggests the following configurations:
- For SDXL and SD3 Medium models: Radeon 7900 series GPUs, Ryzen AI 300 or 395+ series CPUs with 32 GB RAM.
- For SD 3.5 Large and Turbo models: Radeon PRO W7900 or Ryzen AI MAX+ 395 with 64 GB–128 GB RAM.
Built for the Future of Local AI
AMD’s commitment to enabling local AI processing addresses growing privacy concerns and dependence on cloud services. The company’s deep engineering partnership with Stability AI reflects a shared vision of democratized, high-performance AI.
Amuse 3.0 and the corresponding Adrenalin 24.30.31.05 Preview Driver (or the upcoming Adrenalin 24.5.1 mainline) are now available for download. Amuse 3.0 is a beta software and may contain experimental features or stability issues. Performance may vary depending on system configuration.
